The Retirement Consumption Puzzle: Evidence from a Regression Discontinuity Approach